changefeed (ticdc): don't set changefeed to failed state when error r…

…etry reach limit (pingcap#9386)

ref pingcap#5647, close pingcap#8382, close pingcap#9388

cdc/owner/feed_state_manager.go

@@ -463,7 +463,7 @@ func (m *feedStateManager) handleError(errs ...*model.RunningError) {
 			log.Warn("changefeed will not be restarted because it has been failing for a long time period",
 				zap.Duration("maxElapsedTime", m.errBackoff.MaxElapsedTime))
 			m.shouldBeRunning = false
-			m.patchState(model.StateFailed)
+			m.patchState(model.StateError)
 		} else {
 			log.Info("changefeed restart backoff interval is changed",
 				zap.String("namespace", m.state.ID.Namespace),

cdc/owner/feed_state_manager_test.go

@@ -204,7 +204,7 @@ func TestHandleError(t *testing.T) {
 
 	require.False(t, manager.ShouldRunning())
 	require.False(t, manager.ShouldRemoved())
-	require.Equal(t, state.Info.State, model.StateFailed)
+	require.Equal(t, state.Info.State, model.StateError)
 	require.Equal(t, state.Info.AdminJobType, model.AdminStop)
 	require.Equal(t, state.Status.AdminJobType, model.AdminStop)
 

pkg/txnutil/gc/gc_manager.go

@@ -17,7 +17,6 @@ import (
 	"context"
 	"time"
 
-	"github.com/pingcap/errors"
 	"github.com/pingcap/failpoint"
 	"github.com/pingcap/log"
 	"github.com/pingcap/tiflow/cdc/model"
@@ -113,11 +112,7 @@ func (m *gcManager) CheckStaleCheckpointTs(
 ) error {
 	gcSafepointUpperBound := checkpointTs - 1
 	if m.isTiCDCBlockGC {
-		pdTime, err := m.pdClock.CurrentTime()
-		// TODO: should we return err here, or just log it?
-		if err != nil {
-			return errors.Trace(err)
-		}
+		pdTime, _ := m.pdClock.CurrentTime()
 		if pdTime.Sub(oracle.GetTimeFromTS(gcSafepointUpperBound)) > time.Duration(m.gcTTL)*time.Second {
 			return cerror.ErrGCTTLExceeded.GenWithStackByArgs(checkpointTs, changefeedID)
 		}